Agent Skills: Functions + OpenTelemetry

Functions + OpenTelemetry

Critical rules

• Observability is an observer: wrap with trace(); business functions stay telemetry-blind and return Results.
• Structured JSON logs only (Pino) — never string interpolation. Redact sensitive fields in logs and span attributes.
• Use OpenTelemetry semantic attribute names (user.id, not userId).
• Map Result ok/err to span status. Correlate logs with traceId/spanId. Prefer one canonical wide event per request.
• Tests must not change when tracing is off (trace() is a no-op).
• Before wrapping or configuring, read references/patterns.md and references/setup.md.

Workflow

1. Init autotel + Pino with redaction and optional attribute filter — see references/setup.md.
2. Wrap fn(args, deps) with trace((ctx) => async (args, deps) => ...).
3. Set semantic attributes; map Result to span status; never put retry/branch decisions in the core function for telemetry.
4. Ensure logs include trace/span IDs; enable canonical wide events for request roots when useful.
5. Use track() for point-in-time business events; keep existing unit tests unchanged.

Resources

• references/patterns.md — structured logs, trace wrap, semantics, redaction, wide events. Read when instrumenting.
• references/setup.md — init, backends, track vs trace, test no-op. Read when bootstrapping.

Validation

• [ ] Structured fields, not string interpolation
• [ ] Tracing in wrapper only; function stays blind
• [ ] OTel semantic attribute names
• [ ] PII redacted in logs and spans
• [ ] Result maps to span status; logs carry trace/span IDs
• [ ] Existing tests pass unchanged

Constraints

• Not a substitute for proper Result errors. Avoid hot loops where span overhead dominates.
• Related: result-types, fn-args-deps, api-design, resilience.